## C.2 Uniparental Teaching with Transmission Errors

This model is identical to the uniparental teaching model in the main text except that
teaching has an error rate, $\epsilon$, which is the frequency with which a mother pays a teaching
cost, but her offspring fails to learn the behavior. Equation 2 is replaced by Equation C.6
and Equation 3 is replaced by Equation C.7.

$$x''_{sk1} = \frac{1}{2} \left( \underbrace{(τ_{sk} - ϵ)x'_{fk1}}_{\text{Allele from mother}} + \underbrace{(x'_{mk0} + x'_{mk1}) \sum_h (τ_{sh} - ϵ)x'_{fh1}}_{\text{Allele from father}} \right) \quad (C.6)$$

$$x''_{sk0} = \frac{1}{2} \left( \underbrace{x'_{fk0} + (1 - τ_{sk} + ϵ)x'_{fk1}}_{\text{Allele from mother}} + \underbrace{(x'_{mk0} + x'_{mk1}) \sum_h (x'_{fh0} + (1 - τ_{sh} + ϵ)x'_{fh1})}_{\text{Allele from father}} \right) \quad (C.7)$$

As in the main text, I ran numeric simulations of this system for various combinations of teaching costs, *t*, and reproductive benefits of the cultural trait, *b*, and an innovation rate of *r* = 0.005. Since more than 90% of females learn the trait from their mother I conservatively set *ϵ* to 0.1 as the error rate for the simulation. As in the main text, for each parameter combination, I ran the simulation starting with four initial allele frequencies. In each frequency one allele started at 5% of the population and the rest were evenly distributed among the rest of the population. I started the frequency of the cultural trait at zero in all four initial conditions and ran the simulations until they converged to a shared equilibrium or ran for 10⁷ generations.